TiSeLaC ECML/PKDD 2017 discovery challenge solution
This repo hosts the first-place solution to the discovery challenge on Time Series Land cover Classification (TiSeLaC), organized in conjunction of ECML-PKDD 2017.
The challenge consists in predicting the Land Cover class of a set of pixels given their image time series data acquired by the satellites. We propose an end-to-end learning approach leveraging both temporal and spatial information and requiring very little data preprocessing and feature engineering.
The architecture---ranked first out of 21 teams---comprises different modules using dense multi-layer perceptrons, one-dimensional dilated convolutional and fully connected one-dimensional convolutiona neural layers.
To run, the following libraries are required:
- numpy (1.10+),
- sklearn (0.18+),
- keras (2.0+) (we used Theano as a backend).
To train on the full training data, and predict for the whole test, you can run:
ipython -- deep-tsc.py
The code reproduces the obtained results (collected in baML.txt) as reported in the following paper:
Nicola Di Mauro, Antonio Vergari, Teresa M.A. Basile, Fabrizio G. Ventola, Floriana Esposito
End-to-end Learning of Deep Spatio-temporal
Representations for Satellite Image Time Series Classification,
In: Proceedings of the ECML/PKDD Discovery Challenges, 2017